meta-analysis of the theories of reasoned action and

Meta-Analysis of the Theories of Reasoned Action and Planned Behavior 1

Running head: Meta-Analysis of the Theories of Reasoned Action and Planned Behavior

A Meta-Analytic Review of the Theories of Reasoned Action and Planned Behavior in Physical

Activity: An Examination of Predictive Validity and the Contribution of Additional Variables

Martin S. Hagger,

University of Essex,

Nikos L.D. Chatzisarantis,

Brunel University,

Stuart J. H. Biddle,

Loughborough University

Journal of Sport and Exercise Psychology, 24, 1, 3-32

Address for correspondence: Martin S. Hagger Department of Psychology University of Essex Wivenhoe Park Colchester Essex CO4 3SQ United Kingdom Tel: ++44 (0)1206 874898 Fax: ++44 (0)1206 874898 Email: [email protected]


Abstract

The aim of the present study was to examine relations between behavior, intentions, attitudes,

subjective norms, perceived behavioral control, self-efficacy and past behavior across studies

using the Theories of Reasoned Action (TRA) and Planned Behavior (TPB) in a physical activity

context. Meta-analytic techniques were used to correct the correlations between the TRA/TPB

constructs for statistical artefacts across 72 studies and path analyses were conducted to examine

the pattern of relationships among the variables. Results demonstrated that the TRA and TPB

both exhibited good fit with the corrected correlation matrices, but the TPB accounted for more

variance in physical activity intentions and behavior. In addition, self-efficacy explained unique

variance in intention and the inclusion of past behavior in the model resulted in the attenuation of

the intention-behavior, attitude-intention, self-efficacy-intention and self-efficacy-behavior

relationships. There was some evidence for the moderation of the study relationships by attitude-

intention strength and age, but there was a lack of homogeneity in the moderator groups. It was

concluded that the major relationships of the TRA/TPB were supported in this quantitative

integration of the physical activity literature and the inclusion of self-efficacy and past behavior

are important additions to the model.


A Meta-Analytic Review of the Theories of Reasoned Action and Planned Behavior in Physical

Activity: An Examination of Predictive Validity and the Contribution of Additional Variables

Introduction

Acknowledging the strong evidence in favour of regular physical activity for positive health

benefits (Bouchard, Shepherd & Stephens, 1994), exercise psychologists have sought to model

the decision-making processes that lead people to participate in regular physical activity

(Brawley, 1993a). Such an endeavour may identify the key psychological constructs to be

targeted for social-cognitive interventions that promote physical activity participation (Brawley,

1993b). Social cognitive models have been shown to be useful to this end (Maddux, 1993) and

the Theories of Reasoned Action (TRA; Ajzen and Fishbein, 1980) and Planned Behavior (TPB;

Ajzen 1985) have been identified as particularly useful social cognitive theoretical frameworks

that help explain people’s physical activity behavior (Blue, 1995; Godin, 1994).

Hausenblas, Carron and Mack (1997) conducted a cumulative research synthesis of the TRA

and TPB in a physical activity context using meta-analytic techniques. Such analyses treat each

individual study as a single test of the theory relationships and provide a powerful assessment of

the effect sizes across studies by correcting for the statistical artefacts of sampling and

measurement error. The study supported the utility and applicability of the TRA and TPB in the

explanation of physical activity intentions and behavior. However, there is a need to investigate

additional issues raised in the growing body of literature using these social cognitive approaches

in a physical activity context. Specifically, the predictive and construct validity of the

arrangement of the theory constructs, the influence of other social cognitive variables and the

influence of moderators are issues that have not been addressed in previous quantitative research

syntheses of the TRA/TPB in physical activity and requires further investigation (Spence, 1999).


The present study aimed to extend, using meta-analysis and path analytic techniques, the research

of Hausenblas and coworkers (1997) by examining the construct and predictive validity of the

TRA/TPB in physical activity research and evaluating the role of self-efficacy and past behavior

in these models. A sub-aim was to assess the efficacy of three moderator variables, namely

attitude-intention strength, age and time frame of past behavior on the TRA/TPB relationships.

The TRA hypothesizes that an individual’s stated intention to perform a given behavior is

the most immediate predictor of that behavior (Ajzen & Fishbein, 1980). Intention is proposed to

represent a person’s immediate behavioral orientation towards performing a given target

behavior, like physical activity, and is a reflection of a person’s motivation towards that

behavior. The intention-behavior link is a strong one and this is supported by meta-analytic

reviews (Randall & Wolff, 1994; Sheppard, Hartwick, & Warshaw, 1988). The TRA posits that

the intention construct completely mediates the effect of two social cognitive variables on

behavior; attitude and subjective norm. Attitude reflects a personal disposition towards engaging

in the behavior. It represents an individual’s assessment of their beliefs regarding the target

behavior’s agency in producing outcomes and an evaluation of these outcomes. Subjective norm

is a normative-based cognition and represents the person’s evaluation of whether significant

others want them to engage in the target behavior and their motivation to comply with these

others. Ajzen and Fishbein (1980) propose that the TRA is a versatile theory and is applicable to

all volitional human behaviors. This notion has been confirmed on numerous occasions (Randall

& Wolff, 1994; Sheppard et al., 1988).

However, recognising that not all behaviors are under complete volitional control of the

individual, Ajzen (1985) modified the theory by including the perceived behavioral control

(PBC) construct as an independent predictor of intention. PBC represents an individual’s


assessment of their capacities and faculties regarding their behavioral engagement. For example,

some behaviors may have ‘problems of control’ in which case attitude and subjective norm,

while favourable, may not contribute the behavioral intention because of lack of volitional

control. In such cases PBC may independently predict intention. Ajzen and Madden (1986) also

proposed a second version of the TPB in which PBC may also directly influence behavior

independent of the intention-mediated effect. This unique PBC-behavior effect was supposed to

reflect the degree to which actual barriers to the behavior influence enactment of intentions in

which case the PBC measure is a proxy measure of actual control over behavior. Ajzen and

Madden (1986) argue that, for some behaviors, PBC can predict both intention and behavior

because it reflects resources and opportunities as well as actual barriers.

Researchers comparing the TPB and TRA in a physical activity context have demonstrated

that the TPB is superior to the TRA in accounting for the variance in intention. This is because

PBC has been shown to have a large effect on physical activity intentions in the TPB. Indeed,

meta-analytic and narrative reviews of the use of the TRA and TPB in physical activity behavior

research suggest that PBC has an effect on intentions that is almost as strong as that of attitudes

(Blue, 1995; Godin, 1993; Hausenblas et al., 1997). Conversely, these reviews also suggest that

the contribution of subjective norm to the prediction of intention is typically smaller than the

effect of attitude and PBC. This may be consistent with the notion that participation in physical

activity relies more on personal motivational judgements and is less subject to the influence of

pressure from others (Godin, 1993).

The corrected effect sizes yielded by Hausenblas et al.’s (1997) meta-analysis provided

support of the relationships between the TRA/TPB variables, physical activity intentions and

behavior observed in single tests of the model. However, it is important that these relationships


are not considered in isolation if a valued judgement on the construct and predictive validity of

the TRA and TPB is to be made. Hausenblas et al. acknowledge that these important mediator

variables are not accounted for in zero-order relationships and may give a false impression of the

actual effect sizes. For example, it is possible to observe a significant and substantial effect size

between two variables, such as attitude and behavior. Yet, when other influential mediator

variables are partialled out, like intention, the relationship in question can be attenuated or

extinguished altogether (Spence, 1999). Therefore, if an informed decision is to be made on the

validity of the TRA and TPB in an physical activity context, it is important to apply multivariate

techniques, like regression and path analysis, to the synthesis so that the unique relations between

the variables can be generated with the influential mediators partialled out. Adopting the

techniques of meta-analysis and multivariate analyses like path analysis can assist in providing a

strong case for hypothesised mediator relationships in the TRA/TPB for physical activity

behavior.

In addition, Hausenblas et al. (1997) acknowledged that they did not consider additional

variables1 as potential predictors of intention and behavior in their meta-analysis of the

TRA/TPB in an exercise context. Studies adopting the TRA/TPB for physical activity behavior

have catalogued the independent influence of self-efficacy (Courneya & McAuley, 1994a;

Estabrooks & Carron, 1998; Terry & O’Leary, 1995) and frequency of past behavior (Yordy &

Lent, 1993; Godin, Valois, Shephard & Desharnais, 1987) on intention and behavior. The

inclusion of each of these additional variables increased the explanatory power of the model by

addressing some of the shortcomings of the conceptualisation and measures of the existing

TRA/TPB predictors or served to further explain the arrangement of relationships in the model.


Focussing on the concept of self-efficacy, a number of studies have shown that self-efficacy

provides an additional influence on physical activity intentions and behavior (Dzewaltowski,

1989; Dzewaltowski, Noble & Shaw, 1990; Hagger, Chatzisarantis & Biddle, 2001; Yordy &

Lent, 1993). Conceptually, Ajzen (1991; in press) has aligned the concept of PBC with Bandura’s

(1977) self-efficacy construct. However, close examination of the operationalization of PBC and

its system of underlying beliefs has revealed that there may be more than one control-related

construct with independent effects on the TPB variables. Terry and O’Leary (1995) have shown

that items that have been traditionally used to measure PBC can be classified into internal and

external aspects of control and that these may be responsible for the poor internal consistency

values often reported for this measure. The authors characterized these internal aspects of control

as perceived ability and personal agency over the behavior. This construct was measured by

items that make reference to ‘ability’ and ‘capability’ and was labelled self-efficacy. Conversely,

the external aspects of control were defined as the influence of external barriers on behaviour.

This was measured by items referring to ‘difficulty’ and ‘control’ and was labelled perceived

behavioral control. Recently, Armitage and Conner (1999b) have corroborated this internal-

external distinction by confirming the criterion validity of the control variables using their

underlying beliefs. Since this distinction is now well established and the differential effects of

self-efficacy acknowledged in recent reviews of the TPB (Conner & Armitage, 1998), a

cumulative examination of the influence of this variable within the TPB is timely and will

provide further support for making the distinction between PBC and self-efficacy.

The application of the TRA, the TPB and other related social cognitive approaches have

identified that past behavioral engagement, often measured as frequency of past behavior, has an

important and significant influence on prospective behavior, intention and the other social


cognitive variables in the model (Bagozzi, 1981; Bentler & Speckart, 1981). The inclusion of

past behavior, a term often used synonymously with habit, has typically revealed three effects on

the variables in the TRA/TPB. Firstly, past behavior significantly influences the variables of

intention, attitude, subjective norm, PBC and behavior. Secondly, the inclusion of past behavior

in multivariate tests of the influence of these cognitions on intentions and behavior tends to

attenuate the influence of attitudes on intention and intentions on behavior (Bagozzi, 1981;

Hagger, Chatzisarantis, Biddle & Orbell, 2001). Thirdly, while the attenuation effect has reduced

the relationships in studies of some health behaviors (Norman, Conner and Bell, 1999), others

have shown that there are unique effects of the TRA/TPB social cognitions on behavior,

independent of past behavioral engagement (Bentler & Speckart, 1981). Such attenuation effects

have been observed in a number of studies that have examined physical activity behavior

(Dzewaltowski et al., 1990; Godin, Valois, Jobin & Ross, 1991; Godin et al.,1993; Hagger,

Chatzisarantis & Biddle, 2001; Hagger, Chatzisarantis, Biddle & Orbell, 2001; Norman & Smith,

1995; Theodorakis, Doganis, Bagiatis & Gouthas, 1991) while others have only observed small

or nil attenuation effects (Sheeran & Orbell, in press; Godin, Vezina & LeClerc, 1989;

Theodorakis, 1992; Theodorakis, Goudas, Bagiatis & Doganis, 1991; Yordy & Lent, 1993). It

seems that taking past behavior into consideration may account for some of the observed effects

of current cognitions regarding physical activity behavior, like attitudes and intentions.

While the attenuation effects of past behavior in the TRA/TPB have been documented in

previous studies, few hypotheses regarding the mechanisms responsible have put forward.

Bagozzi (1981) suggested that the effects of the current cognitions in the TRA on intention and

behavior become spurious with increasing past behavioral engagement. “As habit increases, the

performance of the behavior becomes less one of a rational evaluation of the consequences of the


act and more one of a learned response” (Bagozzi, 1981, p. 625). Therefore, without controlling

for past behavior, the decision making process is biased towards current judgements regarding

the behavior when, in reality, the decision to engage in the behavior will be less subject to

cognitive evaluation and more one of habitual participation on presentation of the appropriate

cues to action. It was surmised that the extent of past behavioral engagement dictated the degree

of cognitive intervention required before the behavior could be performed, or its level of

‘automaticity’ (Aarts & Dijksterhuis, 2000; Triandis, 1977).

In this way, the role of frequency of past behavior may be able to account for some or all of

the conscious elements that are involved in the decision-making process. While some attenuation

effects have been observed, this has not been systematically investigated across studies of

physical activity behavior, nor has the exact contribution of past behavior to the prediction of

intentions and behavior been documented. Examining the effect of previous experience may shed

light on the extent of the attenuation effects observed in other studies when artefacts of sampling

and measurement error are statistically controlled.

Meta-analyses of the TRA/TPB have also identified some potential moderator variables of

the theory relationships (Sheeran & Orbell, 1998; Sheppard et al., 1988). One possible moderator

is the strength of the attitude-intention relationship. Recent conceptualisations of the TPB have

examined how attitude-based intentions tend to result in stronger intention-behavior relationships

than intentions based on subjective norms. Sheeran, Norman and Orbell (1999) suggest that a

person’s action plan to engage in a behavior is more likely to be translated into behavior if it is

based largely on attitudes. Considering the attitude-intention strength as a moderator of the

TRA/TPB variables in a cumulative research synthesis would provide evidence to support


whether studies reporting strong attitude-intention relationships among their participants also

have strong intention-behavior relationships.

In addition to the moderating effect of attitude-intention strength, previous studies have

highlighted that sample demographics may result in different relative contributions of the

TRA/TPB cognitions to the explanation of intention and behavior. Indeed, Ajzen and Fishbein

(1980) suggest that the relative influences in the TRA will change according to behavior and

sample. It would not, therefore, be unreasonable to assume that in a cumulative analysis there

would be a moderating influence of sample characteristics such as gender, age and socio-

economic status. Moderators such as gender and socio-economic status (SES) are precluded from

the present study because few studies segregate their sample by gender and there is lack of

availability of SES data. However, age is typically reported in most studies making it a viable to

conduct a moderator analysis. Previous meta-analyses of the TPB have shown that age moderates

the intention-behavior relationship (Sheeran & Obell, 1998). Further, studies have shown that

physical activity behavior declines with age (Armstrong and McManus, 1994; USDHHS, 1996).

Therefore it is expected that age would moderate the intention-behaviour relationship.

The time period over which past behavior is assessed is another variable that may serve to

moderate the theory relationships in studies of the TRA/TPB that incorporate the past behavior

variable. A distinction must be made between participants who are asked to report their

frequency of past physical activity behavior over a few weeks or one month and those asked to

assess their past behavior over a period of months or even years. Clearly, there is likely to be a

greater abstraction and generalisation of the past behavior the longer the stated period and this is

likely to be reflected in the measure. It has been suggested that recent assessments of behavioral

engagement are likely to correspond better with future behavior and judgements regarding further


participation (Bagozzi & Warshaw, 1992). Indeed, research has suggested that recency of past

behavior is an important element when assessing the influence of past experience in social

cognitive models (Bagozzi & Kimmel, 1995; Hagger, Chatzisarantis, Biddle & Orbell, 2001).

Godin, Valois, Shepherd and Desharnais (1987) concluded that “if exercise has been experienced

quite recently, [subsequent] behavior should be predictable from a knowledge of exercise habits”

(p. 155). It is therefore important that the moderating effects of the proximity of measurement of

past physical activity behavior is examined when studying the prediction of behavior from past

behavior in a cumulative synthesis of the TRA/TPB research.

Hypotheses

The present study aims to evaluate the predictive validity of the TRA and the TPB in

physical activity behavior by applying path analysis to corrected correlations derived from a

meta-analytic cumulation of the available literature. Specifically, is it hypothesized that the TPB

will account for more variance in intention and behavior than the TRA. It is also expected that

the present analysis will corroborate the findings of other reviews (e.g. Blue, 1995; Godin &

Kok, 1996) that attitude and PBC will have the most pervasive influence on physical activity

intentions.

Further, recognising that the additional variables of self-efficacy and past behavior have been

successful in explaining additional variance in physical activity intentions in the TRA/TPB, the

present study aims to examine the contribution of these additional variables across studies.

Specifically, it is expected that self-efficacy will have a significant and unique effect on physical

activity intention and behavior. Further, it is expected that self-efficacy will exhibit discriminant

validity from PBC as Terry and O’Leary (1995) have found.


The inclusion of past behavior as a predictor of all of the TPB variables will help to unravel

the unbiased effects of the attitude, subjective norm, PBC and self-efficacy on intention and

intention on behavior. It is expected that past behavior will attenuate the attitude-intention and

intention-behavior relationships as shown in individual studies (e.g. Bagozzi & Kimmell, 1995;

Yordy & Lent, 1993).

Finally, the present study will examine the role of three moderating variables in the meta-

analytic cumulation of the TRA/TPB studies on physical activity behavior; strength of the

attitude-intention relationship, age and time period of past behavior assessment. In particular, it is

expected that, according to Sheeran et al.’s (1999) findings, that studies that report strong

average attitude-intention relationships will also exhibit strong intention-behavior correlations.

Further, according to Sheeran and Orbell (1998), it is anticipated that age will moderate the

intention-behavior relationship. It is expected that studies that have measured past behavior in

close proximity to the TPB variables will have stronger past behavior-future behavior

relationships because the assessment is likely to be more accurate due to memory salience.

Method

Literature search

An electronic literature search was conducted using the ATLANTES (1980-1996),

HERACLES (1975-1997), MEDLINE EXPRESS (1980-2001), psychINFO (1977-2001),

SPORT Discus (1975-2001) and Web of Science Social Science Citation Index (1980-2001)

databases. Key words used for the search included exercise, physical activity, theory of planned

behavior, theory of reasoned action, attitudes, intentions and past behavior. A manual search was

conducted on pertinent journals, reference lists and conference proceedings available and we also

performed manual searches of the abstracting journals Dissertation Abstracts International and


Psychological Abstracts. In addition, attempts were made to locate “fugitive literature”

(Rosenthal, 1995, p. 185) by contacting authors to request missing correlations and unpublished

data sets. Criteria for inclusion were studies that defined the target behavior as physical activity,

either as leisure time physical activity or more formal forms such as sports training or exercise,

and reporting at least one correlation between constructs derived from the TRA or TPB.

An initial search identified 90 potentially relevant articles that matched the search criteria.

Some studies were rejected because they did not report the necessary correlations between the

TRA/TPB variables or were qualitative in nature. Some studies scheduled for rejection were later

re-instated when the missing correlations were acquired by contacting the author directly.

Eighteen studies were rejected on the basis of the selection process leaving a total of 72 studies to

be included in the final analysis. Since some studies included multiple data sets, the total possible

sample size was 79. Further, the search yielded 12 studies that included measures of self-efficacy

and 25 studies that included measures of frequency of past behavior within the TRA/TPB

framework.

Meta-analytic strategy

The measure of effect size adopted for evaluation in the present study was the average

correlation coefficient across the studies corrected for statistical artefacts. The meta-analytic

strategy reported by Hunter & Schmidt (1990)2 was used to correct the intercorrelations between

the TRA/TPB variables and past behavior for sampling and measurement error.

Assessment of construct validity

The corrected correlation coefficients were used as input for a path analysis model using the

EQS structural equation modelling programme (Bentler, 1989) to test the construct and predictive

validity of the relationships between the TRA and TPB variables and to test the attenuation


effects of past behavior in the model. The models were evaluated using multiple criteria for

goodness-of-fit: The goodness of fit index (GFI), the comparative fit index (CFI) and the normed

fit index (NNFI), all of which should be greater than .95 for adequate fit of the model with the

input matrix (Hu and Bentler, 1999). In addition, the ‘badness of fit index’, the standardised root

mean squared residual (SRMSR), was also used to evaluate the model. SRMSR values of less

than .08 are proposed to be indicative of an acceptable model (Hu and Bentler, 1995).

Evaluation of moderator variables

The evaluation of moderator variables was conducted by dividing the studies into groups

according to the moderator under scrutiny and recalculating the average reweighted correlations,

standard deviations, standard errors, confidence intervals and credibility intervals using the meta-

analytic strategy outlined previously. The attitude-intention relationships strength and age

moderator variables were coded according to criteria set out in other studies. The moderator was

considered to have a significant effect if the average corrected correlation coefficients from the

moderator groups were significantly different. Moderation was further supported if (a) the

credibility intervals of the moderated analysis were narrower than those reported in the overall

meta-analysis and (b) a small or zero overlap of the confidence intervals of the moderated groups

correlations. Hunter and Schmidt (1990) point out that “to the extent that these confidence

intervals do not overlap, we have sharp confirmation of the predicted moderator variable” (p.

438).

To evaluate the moderating effect of attitude-intention strength, the criteria adopted by

Sheeran et al. (1999) were used. Studies with a significant individual sample-reweighted attitude-

intention relationship of .40 or greater were considered to have a ‘strong’ attitude-intention

relationship and therefore base their intentions predominantly on their attitudes. Due to the


dichotomous nature of the evaluation, the consideration of the subjective norm-intention

relationship was not accounted for. However, in the evaluation of this moderator, the difference

in the strength of the subjective norms-intention relationship between the groups was used as an

additional check on the efficacy of the coding system. It was expected that the high attitude-

intention strength group would have a low subjective norm-intention correlation and vice-versa.

The age moderator was classified according to Sheeran and Orbell’s (1998) criteria. The

sample was divided into groups comprised of adolescent and college/University samples and

older samples. This was done on the basis of reported mean age of the sample with consideration

of the standard deviation and range statistics. If the average age of a sample fell within the

adolescent or college/University age bracket (age less than 25 years) but the range was

particularly high or the standard deviation in excess of 5 years, suggesting a large number of

participants were above this age range, then the study was excluded. Similarly, if no age statistics

were provided, the study was excluded from the moderation analysis. To ensure the classification

system was implemented correctly, three raters were employed to code the studies according to

the criteria outlined above and all agreed perfectly as to the classification of the studies into

‘adolescent’ and ‘older’ samples (r = 1.00).

Finally, the moderating variable of time frame of past behavior measure coded as ‘proximal’

or ‘distal’. Proximal studies were those that asked respondents to report the frequency of their

past physical activity behavior in the last 4 weeks while distal studies were those that asked

respondents about their past physical activity behavior over more than 1 month. Some studies did

not specify a time frame and opted for the ‘average’ or ‘typical’ frequency of past behavior.

These studies were excluded from the analysis.


A summary table of the studies included in the meta-analysis is provided in Table 1. The

table shows the study characteristics and the three moderator variable groups; attitude-intention

strength, age and time frame of past behavior measure.

Results

Corrected correlations

Average correlations corrected for sampling and measurement error for the TRA/TPB

variables and the additional variables of self-efficacy and past behavior are presented in Table 2.

Of the theoretically derived relationships, the correlation matrix followed an expected pattern

with the strongest association being between attitude and intention, followed by the intention-

behavior, PBC-intention and the subjective norm-intention relationships. Interestingly, strong

corrected correlations were exhibited between the control and attitude constructs. These variables

are hypothesised to co-vary in the TPB, but a strong association may be indicative of possible

issues of discriminant validity (multicollinearity) or a possible causal relationship as indicated in

some tests of the TPB (Courneya, 1995; Hagger, 1998). The present analysis demonstrated

moderate-to-strong, positive corrected average correlations between self-efficacy and the TPB

variables. Strong associations were also observed between the TRA/TPB variables and past

behavior, the subjective norms-past behavior relationship excepted. An examination of the 95%

confidence interval for this correlation revealed that it included the value of zero. Therefore, the

hypothesis of a non-zero value for this association in the present sample of studies had to be

rejected. Only three studies reported correlations between self-efficacy and past behaviour and,

while the average is reported, it must be noted that this does not represent a suitable sample size

for such an analysis and is likely to be biased despite the correction for statistical artefacts

(Hunter & Schmidt, 1990).


The credibility intervals for the average sampling and measurement error corrected

correlations from this set of studies included the value of zero or one for five of the correlations,

indicating that there is a great deal of variation in the correlations across the studies. This is

corroborated by the percentage error variance accounted for by the statistical artefacts corrected

for in this meta-analysis. None of the relationships exceeded the Hunter and Schmidt (1990) 75%

rule. This is indicative of heterogeneity in the distribution of the correlations across the studies

and suggestive of the presence of moderating variables.

Path analysis models

Path analysis models were used to evaluate the study hypotheses relating to the construct and

predictive validity of the TRA, TPB and augmentation of the model including self-efficacy and

past behavior3. A systematic approach to the modelling process was employed to address each

study hypothesis in turn. The correlation matrix of the corrected correlations from the meta-

analysis was used as input to the EQS structural equation modelling program. The models were

assessed using a maximum likelihood estimation method with standard deviations of the

correlations fixed to 1.00 as recommended by Bentler (1989). The results of the model tests are

shown in Table 3.

Firstly, the construct validity of Ajzen and Fishbein’s (1980) TRA was assessed. This model

exhibited good fit with the input correlation matrix from the meta-analysis according to the

multiple criteria adopted (Table 3). As expected, intentions significantly predicted behavior (β =

.51, p < .01), attitudes were the strongest significant predictor of intention (β = .56, p < .01) while

subjective norms had a small but significant influence on intentions (β = .12, p < .01). In this

multivariate test of the TRA relationships, attitude accounted for much of the social influences on


intention. Overall, the TRA model constructs explained 37.27% of the variance in intentions and

26.04% of the variance in behavior.

The second step in the analysis involved testing Ajzen’s (1985) TPB. The estimated model

was similar to the TRA, the only difference being that the input matrix contained correlations

with the PBC variable and the model specified a free path between PBC and intention.

Estimation of this model resulted in a significant increase in the goodness of fit χ2 and a lowering

of the fit indexes indicating a compromise in the adequacy of the model to account for the

corrected correlation matrix (Table 3). The model demonstrated that attitude (β = .40, p < .01)

and PBC (β = .33, p < .01) were the best predictors of intentions. In order to test whether the

inclusion of PBC significantly attenuated the path coefficients in the model, 95% confidence

intervals (CI0.95) were calculated for the standardized path coefficients for the models in steps 1

and 2. Attenuation is supported if the higher bound of the CI0.95 for the lower coefficient does not

overlap with the lower bound of the CI0.95 for the higher coefficient. There was no change in the

magnitude of the intention-behavior relationship, but the inclusion of the PBC significantly

attenuated the attitude-intention relationship. As hypothesised, this model accounted for more

variance in intention than the TRA (44.50%).

A third step in the analysis was to test the second version of the TPB in which PBC

contributed directly to the explanation of behavior (Ajzen & Madden, 1986). To this end, the

model estimated in step 2 was modified by releasing a free parameter between PBC and behavior.

This model indicated a significant improvement in model fit as indicated by a significant

reduction in the goodness of fit χ2 and the marginal increase in the fit indices (Table 3). The

contribution of PBC to behavior was significant (β = .15, p < .01). This version of the TPB


accounted for slightly more variance in behavior (27.41%) compared with the first version in step

2.

A fourth step was to include the influence of self-efficacy in the model. It was hypothesised

that self-efficacy would influence both intentions and behavior as shown in other research (e.g.

Dzewaltowski et al., 1990). In order to test this hypothesis from the meta-analytic data, the

corrected correlations for self-efficacy from the meta-analysis were included in the input matrix

of the path analysis and structural paths freed between self-efficacy and intention and between

self-efficacy and behavior. The resulting model fit the data well (Table 3) and indicated that self-

efficacy was a significant predictor of physical activity intention (β = .28, p < .01) and behavior

(β = .15, p < .01). Examination of the CI0.95 for the coefficients between step 3 and step 4

indicated that the attitude-intention relationship was significantly attenuated as a result of the

inclusion of self-efficacy. This model accounted for 50.30% of the variance in intention and

29.10% of the variance in behavior.

In the fifth and final step, the input matrix included correlations with past behavior and the

estimated model introduced free parameters between past behavior and all of the TPB and

additional variables. This model also demonstrated good fit with the data according to the

multiple criteria adopted for this study, although there was a significant increase in the χ2 value

and a slight lowering of the fit indices (Table 3). As hypothesised, past behavior was a significant

predictor of behavior (β = .55, p < .01), intention (β = .37, p < .01), attitude (β = .39, p < .01), SN

(β = .05, p < .01), PBC (β = .23, p < .01) and self-efficacy (β = .58, p < .01). Most pervasive,

though, were the attenuation effects of past behavior in the model. According to CI0.95 statistics,

the inclusion of past behavior in the model significantly attenuated the intention-behavior,

attitude-intention, self-efficacy-intention and self-efficacy-behavior relationships as


hypothesized. The model constructs accounted for the greatest amount of variance in intentions

(86.60%) and behavior (48.16%) compared with the other models tested. This model is

represented schematically in Fig. 1.

Assessment of moderator variables

The evaluation of moderator variables was performed by conducting separate meta-analyses

on the moderator subgroups indicated in Table 1. Results of the moderation analyses for the

attitude-intention strength, age and time frame of past behavioral measurement moderators are

provided in Table 4. Turning first to the attitude-intention relationship strength moderator, the

average, corrected correlation for the intention-behaviour relationship in the strong attitude-

intention relationship group (rc = .50) was significantly lower than the correlation for the weak

attitude-intention group (rc = .58, z = 5.51, p < .01). Not surprisingly the average attitude-

intention correlation was also significantly higher in the strong attitude-intention group.

The examination of age as a moderator of the TPB relationships indicated that older samples

had significantly stronger relationships between intentions and behavior (rc = .57) than younger

samples (rc = .48, z = 8.98, p < .01). This provides some evidence that the older samples may be

more likely to translate their intentions to participate in physical activity into actual behavior.

Finally, there were no significant differences in the corrected correlations between past-future

behavior for the time frame of past behavior moderator groups.

A caveat to these results is that the confidence and credibility intervals of all the effect sizes

in the moderation analysis exhibited substantial overlap and the percentage variance accounted

for by statistical artefacts (sampling and measurement error) indicated that the moderator groups

were not homogenous. This means that there may well be other moderator variables that explain

the residual variance in the effect sizes of each group. Although the statistical artefacts accounted


for the majority of the error variance in the low attitude-intention strength group (64.26%), it did

not surpass the critical 75% value proposed by Hunter and Schmidt (1990) for a homogenous

group of studies. This suggests that the criteria for classifying the groups did not produce the

homogenous groups as expected. Therefore, even though these results are suggestive of a

moderating effect, they must be treated with caution as a significant difference in the average

correlations do not alone confirm moderation (Hunter and Schmidt, 1990).

Discussion

The main focus of the present study was to provide a meta-analytic cumulation of the

studies that have adopted the TRA/TPB as a framework for explaining physical activity

intentions and behavior. The present study aimed to extend the findings of Hausenblas et al.

(1997) by providing path analytic models of the theoretical relationships among the TRA/TPB

variables. This was intended to provide a test of the effectiveness of each model to explain

physical activity behavior. The present study also made a unique contribution to previous

research by examining the role that the external variables of self-efficacy and past behavior have

in the TPB and evaluating the influence of three moderating variables, attitude-intention strength,

age and time frame of past behavior measurement.

An examination of the average correlations corrected for sampling and measurement error

in the present study corroborates the effect sizes reported by Hausenblas et al. (1997). Of the

hypothesised TRA/TPB relationships, medium to large effect sizes were found for the intention-

behavior, attitude-intention and PBC-intention relationships. A smaller effect was shown for the

subjective norm-intention relationship. Substantial non-zero average corrected correlations were

exhibited between all the TRA/TPB variables with the notable exceptions of the relationships of

subjective norm with behavior, PBC and past behavior and the past behavior-PBC relationship. It


was expected that many of these significant zero-order corrected correlations would be mediated

in multivariate tests of the TRA/TPB using path analysis.

A path analysis using the average corrected correlations derived from the meta-analysis

supported the TRA. Intention was the only predictor of behavior and completely mediated the

influence of attitudes and subjective norms on behavior. Intentions were predominantly a

function of attitudes with a small contribution from subjective norms. According to narrative

reviews, the majority of studies in using the TRA/TPB in physical activity behavioral research

have shown that attitudes have the most pervasive influence on intentions and also confirm the

complete mediation of attitudes on behavior by intention (Godin, 1993; Blue, 1995). Further, the

finding that subjective norm had a peripheral influence on intention corroborates the findings of

Hausenblas et al. (1997) who noted that “attitude was over two times more useful as a predictor

of intention than subjective norm” (p. 43). Testing the version of the TPB with only an indirect

PBC-behavior path showed that the PBC variable had a significant influence on intention. This

model exhibited better fit with the input corrected correlation matrix and accounted for more

variance in intention, but not behavior, than the TRA. Such a finding supports the conclusions

put forward by Hausenblas et al. (1997) that the TPB is superior to the TRA, a notion that is also

corroborated by individual research articles comparing the theories in the physical activity

domain (Dzewaltowski et al., 1990; Kimiecik, 1992; Yordy & Lent, 1993). These results also

confirm the conclusion by other reviewers that the contribution of PBC to intentions is as

substantial as that of attitude (Blue, 1995; Hausenblas et al., 1997). However, the present results

show across a number of studies in a physical activity context, that the influence of attitude on

intention is significantly attenuated by the inclusion of PBC. Such attenuation has been observed

in some single empirical studies in a physical activity context (Ajzen & Driver, 1992; Godin et


al., 1989; Kimiecik, 1992) but not in others (Courneya & Friedenreich, 1997, 1999; Courneya et

al., 2000; Kerner & Grossman, 1998). This highlights the importance of a multivariate approach

in a cumulative synthesis of research to show the consistency of such attenuation effects once

study bias has been removed. These findings suggest that attitudes may have some aspects that

are directly related to PBC. A possible reason for this is that the beliefs that underpin attitudes

and PBC may exhibit some conceptual overlap which is reflected in participants responses to

measures of these constructs (Trafimow & Duran, 1998).

Testing the version of the TPB that included a direct PBC-behavior path revealed a

significant direct effect of PBC on behavior. This model also had superior fit with the input

correlation matrix and explained more variance in behavior. It seems that the version of the TPB

in which PBC has dual effects on behavior, direct and indirect via the mediation of intention, is

valid for this sample of studies in a physical activity context (Ajzen & Madden, 1986). It follows

that a person’s intention to participate in physical activity is a function of their attitudinal

evaluation of their future physical activity behavior followed by the degree of subjective control

they have over their ability to engage in physical activity. Further, these findings indicate that

some aspects of control, reportedly those related the barriers and external constraints over the

behaviour (Terry & O’Leary, 1995), are responsible for the enactment of behavior directly. Such

influences transcend the purely cognitive influences of PBC on intention, in which case, PBC

acts as a ‘proxy’ measure of actual control (Ajzen, 1985, 1991). Importantly, this cumulative

synthesis does not make the distinction, as others have (Armitage & Conner, 1999b), between

internal and external aspects of perceived control. This is because this distinction is relatively

recent and most studies tend to confound the two aspects of control by including items measuring

the internal and external aspects of control in a single scale. This has partly been addressed in the


present study by examining the influence of self-efficacy in the TRA/TPB. The self-efficacy

variable makes the internal-external distinction in the present analysis possible, in part, because

self-efficacy is expected to account for the effects that PBC has on intentions and behavior that

are due to internal aspects of control.

The significant influence of self-efficacy on intentions supports the majority of findings in

individual physical activity research articles that have augmented the TPB to include self-

efficacy (e.g. Courneya & McAuley, 1994a; Dzewaltowski et al., 1990; Estabrooks & Carron,

1998; Hagger, Chatzisarantis & Biddle, 2001; Terry & O’Leary, 1995; Van Ryn, Lytle &

Kirscht, 1996). In addition, the fact that the inclusion of self-efficacy has the effect of attenuating

the influence of PBC on intentions, suggests that PBC does indeed have some features that are

largely due to internal aspects of control. This additional variance may be an artefact of

measurement overlap and more precise measures of PBC and self-efficacy that focus clearly on

the internal and external distinction may eliminate this attenuation effect. These results highlight

the importance of both aspects of control to the explanation of physical activity intentions.

These results suggest that interventions to promote increased physical activity behavior in a

population should focus on the promotion of a positive attitude as well as on fostering positive

control over physical activity situations, particularly internal perceptions of control or self-

efficacy. Indeed, research has suggested that attitude and PBC may interact or influence each

other (Courneya, 1995). This indicates that encouraging people to participate in a physical

activity at which they feel competent may promote perceived control over that behavior as

suggested by Skinner (1995). This may, in turn, be considered when a person evaluates their

attitudes towards performing physical activity in the future.


The attenuation effects of past behavior demonstrated in individual studies with the

TRA/TPB (e.g. Godin et al., 1993; Hagger, Chatzisarantis, Biddle & Orbell, 2001; Norman &

Smith, 1995; Yordy & Lent, 1993) is corroborated by the inclusion of past behavior as an

additional variable in present path analysis. When the average corrected correlations of the TPB

constructs with past behavior were included in the input matrix for the path analysis of the TPB,

the intention-behavior, attitude-intention, self-efficacy-intention and self-efficacy-behavior

relationships were significantly attenuated. This suggests that the effects of intentions on

behavior and attitude on intentions observed in studies that do not control for past behavior are

spuriously high.

Despite this attenuation, attitudes remain a pervasive influence on physical activity

intentions even when the effects of past behavior are partialled out, supporting the findings of

Hagger, Chatzisarantis, Orbell and Biddle (2001). This indicates that the immediate decision-

making cognitions prior to forming intentions to participate in physical activity are important as

is an established pattern of physical activity behavior in the past. It seems that while past

behavior has a significant direct influence on intention, attitude, PBC and self-efficacy, these

cognitions are also necessary to translate past decisions about behavioral involvement into action.

This is consistent with the notion that involvement in volitional behaviors like regular physical

activity involves both conscious and automatic influences (Bargh & Chartrand, 1999). Indeed,

there may be elements to people’s attitudes and control that are automatically activated upon

presentation of stimuli. These help people to arrive at their decisions more quickly. However, the

immediate weighing up of the positive and negative aspects of the current attributes of the

upcoming physical activity is also instrumental in determining physical activity behavior.

Practically, then it is important for those involved with organising exercise sessions and


promoting regular physical activity, that the immediate environment is one in which the person is

comfortable and in which they feel confident. This will help enhance positive attitudes and

perceived control toward the physical activity context and promote engagement in the activity.

Since past behavior also has an effect on continued participation, it is also important for exercise

promoters to be aware of these automatic processes and to use them in motivating participation in

physical activity.

Assessing the effect of attitude-intention relationship strength as a moderator of the corrected

correlations demonstrated, not surprisingly, that the high group had a stronger attitude-intention

relationship than the low group. However, the present analysis did not support the hypothesis that

studies with strong attitude-intention relationships exhibited a stronger intention-behavior

relationship as found by Sheeran and colleagues (1999). Rather, the results suggest that a strong

attitude-intention relationship is linked to lower intention-behavior relations. It may be that the

moderating effect of the attitude-intention relationship may hold for certain behaviors and under

particular circumstances. Since the present meta-analysis incorporates studies from many

different contexts and with numerous different actions this effect may be confounded, as

indicated by the lack of homogeneity in the moderator groups.

An interesting finding in the present study is the moderation of the intention-behavior

relationship by age. This suggested that older samples were more likely to implement their

intentions than younger samples. This effect has been found in meta-analyses of the TPB with

other behaviors (Sheeran & Orbell, 1998). These authors suggest that the instability of intentions

among younger samples’ intentions or their relative inexperience with the target behavior may

have resulted in this effect. For physical activity behavior, it is possible that both explanations are

valid. Adolescents and children generally exhibit different operationalizations of the TPB than


adults (Atsalakis & Sleap, 1996; Craig et al., 1996) and may have more difficulties in enacting

their intentions due to external compromises on control such a parental influence (Armstrong &

McManus, 1994). Further, younger samples are not likely to have encountered the physical

activity decision-making process as frequently as older samples and therefore they may form

intentions that are more inconsistent with their behavior. For these reasons, intentions are less

likely to reflect behavior and intention-behavior correlations may be lower for such samples.

It was hypothesized that individuals who have performed physical activity more recently in

the past would possess more elaborate decision-making schema in their memory. However,

present results indicate that the a priori hypothesis that the more recent and distal effects of past

behavior would have differential effects on future behavior should be rejected as time frame of

past behavior was not found to moderate the relationship between past physical activity behavior

and future behavior. This is contrary to the findings of Bagozzi and Kimmel (1995) who

demonstrated that recency of past behavior had a more pervasive effect on intentions than

frequency of past behavior. There are two possible reasons for this incongruence. Firstly, the

classification of the studies into ‘proximal’ and ‘distal’ groups to define the moderator variable in

the present analysis may not have been sensitive enough to make a meaningful distinction.

Further, the relative dearth of studies in the moderator groups may have affected the power of the

analysis as indicated by the large overlap in the confidence intervals, which highlights the

inability of moderator to define homogenous groups.

Conclusion

The present analysis suggests that after accounting for statistical artefacts, people’s attitudes

and, to a lesser extent, perceived behavioral control and self-efficacy seem to be the key

influences in forming intentions to participate in physical activity. Practically, this suggests that


interventions based on the enhancement of attitudes toward physical activity may lead to a

concomitant increase in physical activity behavior. This is supported by other researchers who

suggest that intentions based on attitudes have the strongest effect on behavior (Sheeran et al.,

1999). Further studies have demonstrated that attitude-based messages are very effective in

social-cognitive interventions designed to enhance behavior (Armitage & Conner, 1999a). The

influence of PBC is not as pervasive as reported by other reviews (Blue, 1995), but it is an

important predictor of physical activity behavior nevertheless. The addition of self-efficacy

indicates that internal aspects of control have an unique effect on physical activity intention and

behavior. In summary, the substantial independent contributions made by PBC and self-efficacy

to the explanation of intention and behavior suggests that the TPB augmented by self-efficacy

seems to provide the comprehensive account of the social-cognitive influences on physical

activity motivation and participation.

In addition, the present study indicates that past behavior attenuates relationships among the

TPB constructs in a physical activity context, but does not completely remove the effects of

attitudes on intentions, of intentions on behavior or of PBC on behavior. In some individual

studies the effects of past behavior have been seen to negate much of the influence of attitudes,

subjective norms and PBC on intentions and intentions on behavior. This may precipitate concern

that cognitions like attitudes and subjective norms may not have utility in motivating behavior in

those who have elaborate past experience. However, the most complete demonstration of these

effects have been observed for behaviors that are less subject to volitional control (e.g. smoking)

than physical activity (Norman et al., 1999). Further, there has been no systematic attempt to

control for the statistical artefacts across studies. The present study has controlled for such

artefacts and demonstrated that while past behavioral effects did attenuate the TPB relationships,


current cognitions, particularly control and self-efficacy, are the most important predictors of

physical activity behavior. Future studies in the physical activity domain adopting the TPB as a

framework would do well to account for past physical activity behavior in their analyses in order

to examine the unique influences of conscious social cognitions on intentions and behavior.


References

Aarts, H., & Dijksterhuis, A. (2000). Habits as knowledge structures: Automaticity in

goal-directed behavior. Journal of Personality and Social Psychology, 78, 53-63.

Ajzen, I. (1985). From intentions to actions: A theory of planned behavior. In J. Kuhl & J.

Beckmann (Eds.), Action-control: From cognition to behavior (pp. 11-39). Heidelberg: Springer.

Ajzen, I. (1991). The Theory of Planned Behaviour. Organizational Behavior and Human

Decision Processes, 50, 179-211.

Ajzen, I. (in press). Perceived behavioral control, self-efficacy, locus of control, and the

Theory of Planned Behavior. Journal of Applied Social Psychology.

Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior.

New Jersey: Prentice Hall.

Ajzen, I., & Madden, T. J. (1986). Prediction of goal-directed behavior: Attitudes,

intentions, and perceived behavioral control. Journal of Experimental Social Psychology, 22,

453-474.

Armitage, C., & Conner, M. (1999a, September). Reducing fat intake: Efficacy of a large-

scale intervention based on the theory of planned behaviour. Paper presented at the BPS Division

of Health Psychology Annual Conference, Leeds, UK.

Armitage, C. J., & Conner, M. (1999b). The theory of planned behavior: Assessment of

predictive validity and 'perceived control'. British Journal of Social Psychology.

Armstrong, N., & McManus, A. (1994). Children's fitness and physical activity: A

challenge for physical education. British Journal of Physical Education, 25, 20-26.


Atsalakis, M., & Sleap, M. (1996). Registration of children in a physical activity

program: An application of the Theory of Planned Behavior. Pediatric Exercise Science, 8, 166-

176.

Bagozzi, R. P. (1981). Attitudes, intentions and behavior: A test of some key hypotheses.

Journal of Personality and Social Psychology, 41, 607-627.

Bagozzi, R.P. & Warshaw, P.R. (1992). An examination of the etiology of the attitude-

behavior relation for goal-directed behaviors. Multivariate Behavioral Research, 27, 601-634.

Bargh, J. A., & Chartrand, T. L. (1999). The unbearable automaticity of being. American

Psychologist, 54, 462-479.

Bentler, P. M. (1989). EQS structural equations program manual. Los Angeles: BMDP

Statistical Software.

Bentler, P. M., & Speckart, G. (1981). Attitudes "cause" behaviors: A structural equation

analysis. Journal of Personality and Social Psychology, 40, 226-238.

Blue, C. L. (1995). The predictive capacity of the theory of reasoned action and theory of

planned behavior in exercise research: An integrated literature review. Research in Nursing and

Health, 18, 105-121.

Bouchard, C., Shepherd, R.J., & Stephens, T. (1994). Physical activity, fitness and health:

International proceedings and consensus statement. Champaign, Il: Human Kinetics.

Brawley, L. R. (1993a). Introduction to the Special Issue: Application of social

psychological theories to health and exercise behavior. Journal of Applied Sport Psychology, 5,

95-98.

Brawley, L. R. (1993b). The practicality of using psychological theories for exercise and

health research and intervention. Journal of Applied Sport Psychology, 5, 99-115.


Carron, A. V., Hausenblas, H. A., & Mack, D. (1999). When a comment is much ado

about little: A reply to Spence. Journal of Sport and Exercise Psychology, 21, 382-388.

Conner, M. & Armitage, C. (1998). Extending the Theory of Planned Behaviour: A

review and avenues for further research. Journal of Applied Social Psychology, 28, 1429-1464.

Glass, G. V. (1976). Primary, secondary, and meta-analysis of research. Educational

Researcher, 22, 175-186.

Godin, G. (1993). The theories of Reasoned Action and Planned Behavior: Overview of

findings, emerging problems and usefulness for exercise promotion. Journal of Applied Sport

Psychology, 5, 141-157.

Godin, G. (1994). The theories of reasoned action and planned behavior: Usefulness for

exercise promotion. Medicine and Science in Sports and Exercise, 26, 1391-1394.

Hausenblas, H. A., Carron, A. V., & Mack, D. E. (1997). Application of the Theories of

Reasoned Action and Planned Behavior to exercise behavior: A meta analysis. Journal of Sport

and Exercise Psychology, 19, 36-41.

Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis. Orlando, FL:

Academic Press.

Howell, D.C. (1992). Statistical methods for psychology. Belmont, CA: Duxbury.

Hu, L., & Bentler, P.M. (1999). Cutoff criteria for fit indexes in covariance structure

analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1-55.

Hunter, J., & Schmidt, F. (1990). Methods of meta-analysis: Correcting error and bias in

research findings. Newbury Park, CA: Sage.

Johnson, B. T., Mullen, B., & Salas, E. (1995). Comparison of three major meta-analytic

approaches. Journal of Applied Psychology, 80, 94-106.


Maddux, J. E. (1993). Social cognitive models of health and exercise behavior: An

introduction and review of conceptual issues. Journal of Applied Sport Psychology, 5, 115-140.

Norman, P., Conner, M., & Bell, R. (1999). The theory of planned behavior and smoking

cessation. Health Psychology, 18, 89-94.

Osburn, H. G., & Callender, J. (1992). A note on the sampling variance of the mean

uncorrected correlation in meta-analysis and validity generation. Journal of Applied Psychology,

11, 115-122.

Randall, D. M., & Wolff, J. A. (1994). The time interval in the intention-behaviour

relationship: Meta-analysis. British Journal of Social Psychology, 33, 405-418.

Rosenthal, R. (1995). Writing meta-analytic reviews. Psychological Bulletin, 118, 183-

192.

Rosenthal, R., & Rubin, D. (1988). Comment: Assumptions and procedures in the file

drawer problem. Statistical Science, 3, 120-125.

Schmidt, F. L., & Hunter, J. E. (1999). Comparison of three meta-analysis methods

revisited: An analysis of Johnson, Mullen, and Salas (1995). Journal of Applied Psychology, 84,

144-148.

Sheeran, P., Norman, P., & Orbell, S. (1999). Evidence that intentions based on attitudes

better predict behaviour than intentions based on subjective norms. European Journal of Social


Sheeran, P., & Orbell, S. (1998). Do intentions predict condom use? Meta-analysis and

examination of six moderator variables. British Journal of Social Psychology, 37, 231-250.


Sheppard, B. H., Hartwick, J., & Warshaw, P. R. (1988). The theory of reasoned action: A

meta-analysis of past research with recommendation and future research. Journal of Consumer

Research, 15, 325-343.

Skinner, E. (1995). Perceived control, motivation and coping. Thousand Oaks, CA: Sage.

Spence, J. C. (1999). When a note of caution is not enough: A comment on Hausenblas,

Carron and Mack and theory testing in meta-analysis. Journal of Sport and Exercise Psychology,

21, 376-381.

Trafimow, D., & Duran, A. (1998). Some tests of the distinction between attitude and

perceived behavioural control. British Journal of Social Psychology, 37, 1-14.

Triandis, H. C. (1977). Interpersonal behavior. Monterey, CA: Brookes/Cole.

USDHHS (United States Department of Health & Human Services) (1996). Physical

activity and health: A report of the Surgeon General. McLean, VA: Harvard University Press.


Appendix

Research articles used in meta-analysis.

Ajzen, I., & Driver, B. L. (1991). Prediction of leisure participation from behavioral,

normative, and control beliefs: An application of the theory of planned behavior. Leisure

Sciences, 13, 185-204.

Ajzen, I., & Driver, B. L. (1992). Application of the theory of planned behavior to leisure

choice. Journal of Leisure Research, 24, 207-224.

Amato-Vealey, E. J. (1992). Determinants of exercise behavior after a myocardial

infarction: Beliefs, intention, behavior. Unpublished doctoral dissertation, University of Rhode

Island.

Autrey, P. S. (1999). The determinants of exercise behavior: An application of the

theories of planned behavior and the five-factor model of personality. Unpublished doctoral

dissertation, University of Alabama, Birmingham.

Bagozzi, R. P., & Kimmel, S. K. (1995). A comparison of leading theories for the

prediction of goal directed behaviours. British Journal of Social Psychology, 34, 437-461.

Biddle, S., Goudas, M. & Page, A. (1994). Social-psychological predictors of self-

reported actual and intended physical activity in a University workforce sample. British Journal

of Sports Medicine, 28, 160-163.

Bozionelos, G., & Bennett, P. (1999). The theory of planned behaviour as predictor of

exercise. Journal of Health Psychology, 4, 517-529.

Brenes, G. A., Strube, M. A., & Storandt, M. (1998). An application of the Theory of

Planned Behavior to exercise among older adults. Journal of Applied Social Psychology, 28,

2274-2290.


Brickell, T. (2000). Exercise motivation and levels of participation: A theory of planned

behaviour and self-determination approach. Unpublished Master's Thesis, University of Southern

Queensland, Australia.

Chatzisarantis, N. L. D., & Biddle, S. J. H. (1998). Functional significance of

psychological variables that are included in the theory of planned behaviour: A self-

determination theory approach to the study of attitudes, subjective norms, perceptions of control

and intentions. European Journal of Social Psychology, 28, 303-322.

Chatzisarantis & Biddle (in preparation). A longitudinal study of the intention-behaviour

relationship. Loughborough University, UK.

Chatzisarantis, N. L. D., Biddle, S. J. H., & Meek, G. A. (1997). A self-determination

theory approach to the study of intentions and the intention-behaviour relationship in children's

physical activity. British Journal of Health Psychology, 2, 343-360.

Chow, C.K. & Lindner, K. J. (2001). Physical activity participatory behavior of Hong

Kong adolescents: A test of Ajzen's Theory of Planned Behavior. In A. Papaioannou, M. Goudas,

& Y. Theodorakis (Eds.), Proceedings of the 10th World Congress of the International Society of

Sport Psychology (Vol. 1, pp. 78-80). Skiathos, Greece: Christodoulidi Publications.

Courneya, K. S. (1995). Understanding readiness for regular physical activity in older

individuals: An application of the Theory of Planned Behavior. Health Psychology, 14, 80-87.

Courneya, K. S., Bobick, T. M., & Schinke, R. J. (1999). Does the theory of planned

behavior mediate the relation between personality and behavior? Basic and Applied Social



Courneya, K. S., & Friedenreich, C. M. (1997). Determinants of exercise during

colorectal cancer treatment: An application of the theory of planned behavior. Oncology Nursing

Forum, 24, 1715-1723.

Courneya, K. S., & Friedenreich, C. M. (1999). Utility of the theory of planned behavior

for understanding exercise during breast cancer treatment. Psycho-Oncology, 8, 112-122.

Courneya, K. S., Friedenreich, C. M., Arthur, K., & Bobick, T. M. (1999). Understanding

exercise behavior in colorectal cancer patients: A prospective study using the theory of planned

behavior. Rehabilitation Psychology, 44, 68-84.

Courneya, K. S., & McAuley, E. (1994a). Are there different determinants of the

frequency, intensity, and duration of physical activity? Behavioral Medicine, 20, 84-90.

Courneya, K. S., & McAuley, E. (1994b). Factors affecting the intention-physical activity

relationship: Intention versus expectation and scale correspondence. Research Quarterly for

Exercise and Sport, 65, 280-285.

Courneya, K. S., & McAuley, E. (1995). Cognitive mediators of the social influence-

exercise adherence relationship: A test of the Theory of Planned Behavior. Journal of

Behavioural Medicine, 18, 499-515.

Courneya, K. S., Nigg, C.R., & Estabrooks, P. A. (1998). Relationships among the theory

of planned behavior, stages of change, and exercise behavior in older persons over a three year

period. Psychology and Health, 13, 355-367.

Courneya, K. S., Plotnikoff, R. C., Hotz, S. B., & Birkett, N. J. (2000). Social support and

the theory of planned behavior in the exercise domain. American Journal of Community Health,

24, 300-308.


Craig, S., Goldberg, J., & Dietz, W. H. (1996). Psychosocial correlates of physical

activity among fifth and eighth graders. Preventative Medicine, 25, 506-513.

Daltroy, L. H., & Godin, G. (1989). The influence of spousal approval and patient

perception of spousal approval on cardiac patient participation in exercise programs. Journal of

Cardiopulmonary Rehabilitation, 9, 363-367.

Dzewaltowski, D. A. (1989). Toward a model of exercise motivation. Journal of Sport


Dzewaltowski, D. A., Noble, J. M., & Shaw, J. M. (1990). Physical activity participation:

Social cognitive theory vs. the theories of reasoned action and planned behavior. Journal of Sport


Estabrooks, P. & Carron, A. V. (1998). The conceptualization and effect of control beliefs

on exercise attendance in the elderly. Journal of Aging and Health, 10, 441-457.

Godin, G., Colantonio, A., Davis, G. M., Shephard, R. J., & Simard, C. (1986). Prediction

of leisure time exercise behavior among a group of lower-limb disabled adults. Journal of

Clinical Psychology, 42, 272-279.

Godin, G., & Gionet, J. G. (1991). Determinants of an intention to exercise of an electric

power commission's employees. Ergonomics, 34, 1221-1230.

Godin, G., & Shephard, R. J. (1986a). Importance of type of attitude to the study of

exercise behavior. Psychological Reports, 58, 991-1000.

Godin, G., & Shephard, R. J. (1986b). Psychosocial factors influencing intentions to

exercise of young students from grades 7-9. Research Quarterly for Exercise and Sport, 57, 41-

52.


Godin, G., Valois, P., Jobin, J., & Ross, A. (1991). Prediction of intention to exercise of

individuals who have suffered from coronary heart disease. Journal of Clinical Psychology, 47,

762-772.

Godin, G., Valois, P., & LePage, L. (1993). The pattern of influence of perceived

behavioral control upon exercising behavior: An application of Ajzen's theory of planned

behavior. Journal of Behavioral Medicine, 16, 81-102.

Godin, G., Valois, P., Shephard, R. J., & Desharnais, R. (1987). Prediction of leisure-time

exercise behavior - A path analysis (LISREL V) model. Journal of Behavioral Medicine, 10, 145-

158.

Godin, G., Vezina, L., & LeClerc, O. (1989). Factors influencing the intention of

pregnant women to exercise after birth. Public Health Reports, 104, 185-195.

Greenockle, K. M., Lee, A. A., & Lomax, R. (1990). The relationship between selected

students’ characteristics and activity patterns in a required high school physical education class.

Research Quarterly for Exercise and Sport, 61, 59-69.

Gyurcsik, N. C., & Brawley, L. R. (2000). Mindful deliberation about exercise: Influence

of acute positive and negative thinking. Journal of Applied Social Psychology, 30, 2513-2533.

Hagger, M. S. (1998). The role of perceived control in the Theory of Planned Behaviour

in a physical activity context. Unpublished doctoral thesis, Loughborough University, UK.

Hagger, M. S., Cale, L. & Ashford, B. (1997). Children's physical activity levels and

attitudes towards physical activity. European Physical Education Review, 3, 144-164.

Hagger, M. S., Chatzisarantis, N. & Biddle, S. (in press). The influence of autonomous

and controlling motives on physical activity intentions within the Theory of Planned Behaviour.

British Journal of Health Psychology.


Hagger, M. S., Chatzisarantis, N. & Biddle, S. (2001). The influence of self-efficacy and

past behavior on the physical activity intentions of young people. Journal of Sports Sciences, 19,

711-725.

Hagger, M. S., Chatzisarantis, N., Biddle, S., and Orbell, S. (2001). Antecedents of

children’s physical activity intentions and behaviour: Predictive validity and longitudinal effects.

Psychology and Health, 16, 391-407.

Jackson, C., Smith, R. A., & Conner, M. (1999, September). Does the inclusion of

additional social norm variables to the TPB increase the prediction of physical activity intention?

Paper presented at the 1999 British Psychological Society Division of Health Psychology Annual

Conference, University of Leeds, Leeds, UK.

Kerner, M. S. & Grossman, A. H. (1998). Attitudinal, social, and practical correlates of

fitness behavior: A test of the theory of planned behavior. Perceptual and Motor Skills, 87, 1139-

1154.

Kimiecik, J. (1992). Predicting vigorous physical activity in corporate employees:

Comparing theories of reasoned action and planned behavior. Journal of Sport and Exercise


Legg, D. L. (1986). An investigation of college student exercise behavior using Fishbein's

model: The theory of reasoned action. Unpublished doctoral dissertation, University of Toledo,

OH.

Lowe, R., Eves, F., & Carroll, D. (in press). The influence of affective and instrumental

beliefs on exercise intentions and behavior: A longitudinal analysis. Journal of Applied Sport

Psychology.


Michels, T. C., & Kugler, J. P. (1998). Predicting exercise in older Americans: Using the

Theory of Planned Behavior. Military Medicine, 163, 524-529.

Miller, P., Wikoff, R. L., McMahon, M., Garrett, M. J., & Ringel, K. (1985). Indicators of

medical regimen adherence for myocardial infarction patients. Nursing Research, 34, 268-272.

Mummery, W. K., Spence, J. C., & Hudec, J. C. (2000). Understanding physical activity

intention in Canadian school children: An application of the theory of planned behavior.

Research Quarterly for Exercise and Sport, 71, 116-124.

Norman, P., Conner, M., & Bell, R. (2000). The theory of planned behaviour and

exercise: Evidence for the moderating role of past behaviour. British Journal of Health


Norman, P., & Smith, L. (1995). The theory of planned behaviour and exercise: an

investigation into the role of prior behaviour, behavioural intentions and attitude variability.

European Journal of Social Psychology, 25, 403-415.

Payne, N., Jones, F., & Harris, P. (2000, September). Does the inclusion of additional

social norm variables to the TPB increase the prediction of physical activity intention? Paper

presented at the 2000 British Psychological Society Division of Health Psychology Annual

Conference, University of Kent, Canterbury, UK.

Payne, N., Jones, F., & Harris, P. (2001). Does the inclusion of additional social norm

variables to the TPB increase the prediction of physical activity intention? Manuscript in

preparation, University of Hertfordshire, UK.

Pender, N. J., & Pender, A. R. (1986). Attitudes, subjective norms and intentions to

engage in health behaviors. Nursing Research, 35, 15-18.


Reynolds, K. D., Killen, J. D., Bryson, S. W., Maron, D. J., Barr Taylor, C., Maccoby, N.,

& Farquhar, J. W. (1990). Psychosocial factors of physical activity in adolescents. Preventative

Medicine, 19, 541-551.

Riddle, P. K. (1980). Attitudes, beliefs, behavioral intentions and behaviors of women

and men toward regular jogging. Research Quarterly for Exercise and Sport, 51, 663-674.

Rosen, C. S. (2000). Integrating stage and continuum models to explain processing of

exercise messages and exercise initiation among sedentary college students. Health Psychology,

19, 172-180.

Schmelling, E. C. (1985). Identifying the salient outcomes of exercise: Applications of

marketing principles to preventative health behavior. Public Health Nursing, 2, 93-103.

Sheeran, P., & Orbell, S. (In press). Self-schemas and the theory of planned behaviour.

British Journal of Health Psychology.

Smith, R. A., & Biddle, S. J. H. (1999). Attitudes and exercise adherence: Test of the

theories of reasoned action and planned behaviour. Journal of Sports Sciences, 17, 269-281.

Terry, D. J., & O'Leary, J. E. (1995). The Theory of Planned Behaviour: The effects of

perceived behavioural control and self-efficacy. British Journal of Social Psychology, 34, 199-

220.

Theodorakis, Y. (1992). Prediction of athletic participation: A test of planned behaviour

theory. Perceptual and Motor Skills, 74, 371-379.

Theodorakaris, Y. (1994). Planned behavior, attitude strength, role identity, and the

prediction of exercise behavior. The Sport Psychologist, 8, 149-165.


Theodorakaris, Y., Doganis, G., Bagiatis, K., & Gouthas, M. (1991). Preliminary study of

the ability of Reasoned Action Model in predicting exercise behavior in young children.

Perceptual and Motor Skills, 72, 51-58.

Theodorakis, Y., Goudas, M., Bagiatis, K., & Doganis, G. (1991). Reasoned action theory

and the prediction of training participation in young swimmers. British Journal of Physical

Education, 10, 10-13.

Trafimow, D., & Trafimow, J. H. (1998). Predicting back pain sufferers' intentions to

exercise. Journal of Psychology, 132, 581-592.

Valois, P., Desharnais, R., & Godin, G. (1988). A comparison of the Fishbein and Ajzen

and Triandis attitudinal models for the predication of exercise intention and behavior. Journal of

Behavioral Medicine, 11, 459-472.

Van Ryn, M., Lytle, L. A., & Kirscht, J. P. (1996). A test of the Theory of Planned

Behavior for two health-related practices. Journal of Applied Social Psychology, 26, 871-883.

Wankel, L., Mummery, K. Stephens, T., & Craig, C. (1994). Prediction of physical

activity intention from social psychological variables: Results from the Campbell’s survey of

well-being. Journal of Sport and Exercise Psychology, 21, 213-228.

Warshaw, P. R., & Davis, F. (1985). Disentangling behavioral intention and behavioral

expectation. Journal of Experimental Social Psychology, 21, 213-228.

Yordy, G., & Lent, R. (1993). Predicting aerobic exercise participation: Social cognitive,

reasoned action and planned behavior models. Perceptual and Motor Skills, 76, 287-292.


Note.

1Since researchers have acknowledged that the subjective norms construct only taps a

limited number of a multitude of possible social influences on physical activity intentions, studies

have recently examined the influence of alternative approaches to social factors in the TPB.

Variables such as social support (Courneya & McAuley, 1995; Courneya et al., 2000) and role

beliefs (Bozionelos & Bennett, 1999) have been included in the TPB and have been shown to

explain unique variance in physical activity intentions. However, the limited number of studies

adopting these alternative subjective norms variables in a physical activity context precludes a

meta-analytic cumulation of studies using social support and role beliefs at this stage.

2The Hunter and Schmidt approach is one of three prominent traditions in meta-analysis

that have emerged since Glass’ (1976) classic paper on research synthesis. The two alternative

set of techniques that have received much attention are those put forward by Hedges and Olkin

(1985) and Rosenthal and Rubin (1988). It is important to note that the Hunter and Schmidt

(1990) methods have been criticised for producing anomalous results. For example, Johnson,

Mullen and Salas (1995) report that the Hedges and Olkin (1985) and Rosenthal and Rubin

(1998) approaches confirm that the likelihood of finding the same result due to chance in a

sample of studies decreases as the number of studies included increases, while the Hunter and

Schmidt (1990) technique does not. However, Schmidt and Hunter (1999) have shown that the

differences found in Johnson and colleagues’ (1995) analysis can be accounted for by the latter

researchers’ use of a formula for the standard error of the mean correlation that Hunter and

Schmidt (1990) did not use. Further, Schmidt and Hunter (1999) claim that the use of a fixed-

effects (FE) model is less accurate in computer simulations of population correlations than a

random effects (RE) model. The FE model assumes no real variance in the population


correlations and is favoured by the Hedges and Olkin (1985) and Rosenthal and Rubin (1988)

while the RE model assumes that the sampling error variance varies across studies and is used in

the Hunter and Schmidt (1990) algorithms. Given that Osburn and Callender (1992) contend that

it is improbable that the error variance across real sets of studies is likely to be zero, the use of a

RE model to calculate the mean standard error of the average correlations is the most appropriate

method to adopt.

3Carron, Hausenblas and Mack (1999) report that they did not perform a multivariate

analysis in their meta-analysis (Hausenblas et al., 1997) because they did not have enough

statistical power to justify the analysis. It was therefore important to conduct a power analysis

with the present data to ensure sufficient effect sizes were available to warrant such an analysis.

Assuming large effect sizes between the model variables and setting alpha at .05 and power at

.80, it would be necessary to have 12 effect sizes (correlations) for a meaningful path analysis to

be conducted (see Howell, 1992, p. 210). Table 2 shows that there were sufficient effect sizes

available to conduct a path analysis with the present data set for all of the TRA/TPB

relationships.


Table 1. Study characteristics and grouping data for the three moderator variables.

Authors N Sample composition A-I

correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Ajzen & Driver (1991) 111 College students - - 20.1

(range: 17 to 40)

Younger - -

Ajzen & Driver (1992) 146 College students .44 Strong 20.1

(range: 17 to 40)

Younger - -

Amato-Vealey (1992) 185 Post-myocardial

infarction patients

.37 Weak 61.0

(range: 35 to 80)

Older - -

Autrey (1999) 101 Participants in exercise

program at personal

training centers

.52 Strong Age range 18 to 71 Not

classified

Past 3 months Distal

Bagozzi & Kimmel (1995) 139 Undergraduate students .64 Strong - - - -

Biddle, Goudas & Page (1994) 131 University employees .55 Strong - - - -

Bozionelos & Bennett (1999) 114 Undergraduate students .44 Strong 22.0 (SD=4.88) Younger Past 3 weeks Proximal

Brenes, Strube & Storandt

(1998)

105 Exercise class attendees .13 Weak 68.29 (SD=6.63)

(range: 53 to 84)

Older Frequency in each

decade of life

Not

classified

Brickell (2000) 208 Members and staff of

health clubs

.28 Weak 23.94 (3.92)

(range: 18 to 32)

Younger - -


Table 1. Continued Authors N Sample composition A-I

correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Chatzisaratis & Biddle (1998) 102 Company employees and

manual workers

.89 Strong 39.96 (SD=10.66) Older - -

Chatzisarantis & Biddle (in

preparation)

70 School children .61 Strong - - - -

Chatzisaratis, Biddle & Meek

(1997)

100 School children - - 13.5

(range: 11 to 15)

Younger - -

Chow & Lindner (2001) 923 School children .50 Strong Age range 12 to 15 Younger 2 weeks Proximal

Courneya (1995) 288 Randomly selected

members of social

service center

.89 Strong 71.5 Older - -

Courneya, Bobick & Schinke

(1999) (sample 1)

300 Female undergraduates .58 Strong 19.56 (SD=3.82) Younger - -

Courneya, Bobick & Schinke

(1999) (sample 2)

300 Aerobics class

participants

.46 Strong 25.00 (SD=7.98) Older - -

Courneya & Friedenreich

(1997)

110 Colorectal cancer

survivors

.54 Strong 60.9 (SD=10.8)

(range: 26 to 77)

Older Typical week Not

classified

Courneya & Friedenreich

(1999)

164 Breast cancer patients .41 Strong 53.0 (SD=9.4) Older - -



correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Courneya, Friedenreich, Arthur

& Bobick (1999)

66 Colorectal cancer

patients

.46 Strong 60.8 (SD=11.5) Older Typical week Not

classified

Courneya & McAuley (1994a,

1994b)

170 Undergraduate students .44 Strong 20.34 (SD=2.15) Younger Past 4 weeks Proximal

Courneya & McAuley (1995) 62 Volunteers from

University aerobic

program

.29 Weak 38.79 (SD=14.73) Older - -

Courneya, Nigg & Estabrooks

(1998)

131 Members of social,

recreational and

educational facility

.49 Strong 71.5 (SD=6.0) Older - -

Courneya, Plotnikoff, Hotz &

Birkett (2000)

1557 Randomly selected

residents from Canadian

telephone exchange

.41 Strong 38.97 (SD=11.39) Older - -

Craig, Goldberg & Dietz (1996)

(sample 1 – grade 5 students)

154 Elementary school

children

.43 Strong 10.8 (SD=.05) Young - -

Craig, Goldberg & Dietz (1996)

(sample 2 – grade 8 students)

151 Elementary school

children

.47 Strong 13.9 (SD=.05) Young - -



correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Daltroy & Godin (1989) 132 Cardiac patients on

medically supervised

exercise programs

- - 53.5 (SD=7.6) Older - -

Dzewaltowski (1989) 328 Undergraduate students .48 Strong - - - -

Dzewaltowski, Noble & Shaw

(1990)

254 Undergraduate students .83 Strong - - Past 1 week Proximal

Estabrooks & Carron (1998) 157 Exercise class attendees .07 Weak 68.0 (SD=7.87) Older - -

Godin, Colantonio, Davis,

Shepherd & Simard (1986)

62 Male adults with lower

limb disability

.24 Weak 30.70 Older Frequency per

month

Not

classified

Godin & Gionet (1991) 444 Employees of electric

power commission

.38 Weak 36.3 (SD=8.8) Older Past 6 months Distal

Godin & Shepherd (1986a) 90 University employees .54 Strong Stratified into three

age groups (ranges:

45 to 54, 55 to 64,

65 to 74)

Older Past 4 months Distal

Godin & Shepherd (1986b) 698 High school students .59 Strong 13.0 (SD=1.0) Younger - -



correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Godin, Valois, Jobin & Ross

(1991)

161 Coronary Heart Disease

patients

.42 Strong 52.8 (SD=8.1) Older Past 3 months Distal

Godin, Valois & Lepage (1993)

(study 1)

347 Randomly selected

persons from telephone

directory

.46 Strong Age range 18 to 40 Not

classified


Godin, Valois & Lepage (1993)

(study 2)

136 Postnatal women .54 Strong - - Past 3 months Distal

Godin, Valois, Shepherd &

Desharnais (1987)

129 University employees .62 Strong Stratified into three

age groups (ranges:

less than 30, 30 to

44, 45 and above)

Not

classified


Godin, Vezina & Leclerc (1989) 98 Pregnant women

attending antenatal

classes

.50 Strong 28.6 (SD=3.9) Older Past 12 months Distal

Greenockle, Lee & Lomax

(1990)

136 School children - - 14.7 (SD=.96) Younger - -

Gyurcsik & Brawley (2000) 82 University students - - 23.7 (SD=5.8) Younger - -



correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Hagger (1998) 181 School children .73 Strong Age range 12 to 14 Younger - -

Hagger, Cale & Ashford (1997) 45 School children .24 Weak Age range 9 to 11 Younger - -

Hagger, Chatzisarantis and

Biddle (in press)

1088 School children .51 Strong Age range 12 to 14 Younger - -

Hagger, Chatzisarantis & Biddle

(2001)


Hagger Chatzisarantis, Biddle &

Orbell (2001)


Jackson, Smith & Conner

(1999)

168 Employees of a Higher

Education college

.50 Strong 42.9 (SD=9.36) Older - -

Kerner & Grossman (1998) 73 Members of commercial

fitness center

.38 Weak 44.7 (SD=9.9)

(range: 20.3 to

67.1)

Older - -

Kimiecik (1992) 332 Corporate employees .77 Strong 39.1 (range: 18 to

67)

- - -

Legg (1986) 99 College students .33 Weak Age range 18 to 54 Not

classified

- -



correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Lowe, Eves & Carroll (in press) 369 Patients from four

General Practitioner

patient lists

.58 Strong Males (N=149):

46.4 (SD=15.5);

Females (N=220):

44.2 (SD=15.5)

Older - -

Michels & Kugler (1998) 394 Random sample of

persons aged 65 to 70

years from hospital

registration list

.33 Weak Age range 65 to 70 Older Past 1 week Proximal

Miller, Wikoff, McMahon,

Garrett & Ringel (1985)

141 Patients from cardiac

rehabilitation centres

- - Age range 30 to 70 Older - -

Mummery, Spence & Hudec

(2000) (sample 1 – grade 3

students)

163 School pupils randomly

selected from Canadian

schools

.40 Strong 8.2 (SD=0.5) Younger - -



students)



schools




correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc



students)



schools




students)



schools


Norman, Conner & Bell (2000) 87 Undergraduates .33 Weak - - - -

Norman & Smith (1995) 83 Sample identified by

undergraduate

psychology class

.42 Strong 29.89 (SD=9.27) Older Frequency per

week

Proximal

Payne, Jones & Harris (2000) 241 Employees of a UK

computer company

.33 Weak Stratified into four

age groups (ranges:

16 to 24, 25 to 34,

35 to 44, 45 to 54,

55 or over)

Not

classified

- -



correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Payne, Jones & Harris (2001) 331 Employees of a UK

computer company

32 Weak Stratified into four

age groups (ranges:

<19, 20 to 39, 40 to

59, ≥60)

Not

classified

- -

Pender & Pender (1986) 377 Community residents

randomly selected

.18 Weak 38 (SD=12) Older - -

Reynolds et al. (1990) 374 School children - - 15 (range: 14 to 16) Younger - -

Riddle (1980) 237 Volunteers to research

project

.73 Strong >30 years Older - -

Rosen (2000) 149 Undergraduate students .45 Strong - - Past 1 week Proximal

Schmelling (1985) 135 University employees .77 Strong Age range 30 to 55 Older - -

Sheeran & Orbell (in press) 163 Undergraduate students .61 Strong - - Past 2 weeks Proximal

Smith & Biddle (1999) (sample

1)

96 Members of private

sector health club

.12 Weak 34.0 (SD=12.6) Older - -

Smith & Biddle (1999) (sample

2)

155 Employees of rural

district council in South-

West England

.49 Strong 36.0 (SD=11.6) Older - -



correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Terry & O’Leary (1995) 146 Undergraduate students .65 Strong 20.29 Younger Past 2 weeks Proximal

Theodorakis (1992) 98 Young people attending

swimming

.20 Weak Age range 10 to 13 Younger Past 1 month Proximal

Theodorakis (1994) 395 Female attendees to

fitness clubs

.39 Weak 29.27 (SD=8.75)

(range 18 to 45)

Older - -

Theodorakis, Doganis, Bagiatis

& Gouthas (1991)

56 Children volunteering to

take part in study

.32 Weak Age range 10 to 11 Younger Past 3 years Distal

Theodorakis, Goudas, Bagiatis

& Doganis (1991)

105 Young people attending

swimming

.30 Weak Age range 9 to 12 Younger Past 3 years Distal

Trafimow & Trafimow (1998) 23 Patients diagnosed with

lower back pain

.07 Weak - - - -

Valois, Desharnais & Godin

(1991)

166 University employees - - 39.7 (SD=10.7) Older - -

Van Ryn, Lytle & Kirscht

(1996)

142 Telephone company

employees

.43 Strong 42 (range: 26 to 61) Older - -



correlation

A-I

strengtha

Mean Age (years),

SD & Range

Age

categoryb

Time frame of past

behavior measure

Proximityc

Wankel, Mummery, Stephens &

Craig (1994)

3629 Household-based,

stratified, multistage

cluster survey

.36 Weak Stratified into four

age groups (ranges:

<19, 20 to 39, 40 to

59, ≥60)

Not

classified

- -

Warshaw & Davis (1985) 197 Undergraduate students - - - - - -

Yordy & Lent (1993) 284 Psychology

undergraduates

.76 Strong 19.4 (SD=2.29) Younger Past 4 weeks Proximal

Note. a Moderator group for attitude-intention strength

b Moderator group for age

c Moderator group for proximity of past behavior measure


Table 2. Results of the meta-analysis of the TRA/TPB and past behavior in an exercise context.

Relationship k N rca rc

b Confidence Interval✝

Credibility Interval✝✝

SD SE Variancec

Lower bound

Upper bound

Lower bound

Upper bound

Intention-Behavior 60 10985 .42 .51 .45 .67 .17 .85 .21 .03 15.69

Attitude-Intention 70 20240 .48 .60 .55 .65 .26 .93 .20 .03 12.21

Sub. Norm-Intention 65 19579 .25 .32 .27 .37 .01 .63 .19 .03 13.33

PBC-Intention 49 16732 .44 .57 .51 .63 .25 .90 .19 .03 18.48

PBC-Behavior 35 7173 .31 .39 .33 .45 .12 .66 .17 .03 20.75

Attitude-Sub. Norm 49 13637 .27 .36 .29 .42 .01 .71 .21 .03 11.59

Attitude-PBC 39 11857 .39 .55 .46 .65 .08 1.02 .29 .05 8.51

Sub. Norm-PBC 39 11857 .25 .37 .28 .46 -.06 .80 .26 .04 10.01

Attitude-Behavior 44 8493 .30 .35 .30 .41 .11 .60 .15 .03 23.07

Sub. Norm-Behavior 42 8651 .15 .17 .23 .12 -.07 .43 .15 .03 22.74

Past beh. -Behavior 22 4110 .58 .64 .56 .72 .33 .95 .19 .04 7.55

Past beh. -Intention 25 4727 .48 .58 .51 .66 .31 .86 .17 .04 21.71

Past beh. -Attitude 22 4303 .33 .39 .33 .44 .20 .57 .11 .03 32.91

Past beh. -Sub. Norm 22 4303 .04 .05 -.08 .19 -.45 .56 .31 .07 6.88

Past beh. -PBC 18 3930 .18 .23 .08 .39 -.30 .76 .32 .08 6.96

Self-eff.-Behaviour 9 2402 .36 .40 .34 .46 .28 .52 .07 .03 39.27

Self-eff.-Intention 11 2841 .47 .55 .43 .66 .24 .86 .19 .06 9.77

Self-eff.-Attitude 9 2645 .40 .48 .34 .61 .15 .81 .20 .07 8.33

Self-eff.-Sub Norm 8 2475 .22 .26 .16 .37 .03 .49 .14 .05 17.48

Self-Eff.-PBC 7 2119 .31 .42 .31 .53 .23 .62 .11 .05 28.08

Self-Eff.-Past beh. 3 1353 .39 .45 .33 .57 .29 .61 .19 .06 18.28

Note. k = Number of Studies N = Sample size a Average re-weighted correlation corrected for sampling error only b Average re-weighted correlation corrected for sampling error and measurement error


Note. ✝ 95% Confidence Interval of average re-weighted correlation corrected for sampling error only ✝✝ 90% Credibility Interval of average re-weighted correlation corrected for sampling and measurement error SD = Standard deviation of mean SE = Standard error of mean c Percentage of error variance accounted for by statistical artefacts


Table 3. Results of path analyses tests of the TRA, TPB, TPB augmented with self-efficacy and TPB with past behavior using meta-analytic derived

correlations.

Step χ2 df GFI CFI NNFI SRMSR ∆χ2 I→B A→I SN→I PBC→

I

PBC

→B

SE→I SE→

B

PB→

B

PB→I PB→

A

PB→

SN

PB→

PBC

PB→

SE

1 51.890 2 .998 .996 .987 .014 - .510* .557* .119* - - - - - - - -

2 246.654 3 .992 .986 .952 .028 194.764* .510* .399* .054* .331* - - - - - - -

3 15.574 2 .999 .999 .996 .006 231.080* .426* .399* .054* .331* .147* - - - - - -

4 16.480 3 1.000 .999 .995 .006 .906 .356* .302* .038* .272* .123* .281* .153* - - - - -

5 86.602 2 .998 .997 .972 .009 70.122* .047* .200* .087* .281* .223* .149* .035* .546* .366* .390* .050* .450* .450*

Note. Step 1. Theory of Reasoned Action Step 2. Theory of Planned Behavior Step 3. Theory of Planned Behavior augmented with direct PBC→B path Step 4. Theory of Planned Behavior augmented with PBC→B, SE→I and SE→B paths Step 5. Theory of Planned Behavior augmented to include control for Past Behavior I = Intention B = Behavior A = Attitude SN = Subjective Norms PBC = Perceived Behavioral Control SE = Self-Efficacy PB = Past Behavior * p < .01 ** p < .05


Table 4. Summary of the results of the moderator analyses.

Moderator Relationship k N rca rc

b Confidence Interval✝ Credibility Interval✝✝ SD SE Variancec

Lower

bound

Upper

bound

Lower

bound

Upper

bound

A-I strength I-B 17 (34) 2853 (6721) .39 (.45) .58 (.50) .43 (.43) .72 (.57) .11 (.19) 1.04 (.81) .29 (.19) .07 (.03) 12.81 (11.31)

Att-I 22 (48) 7272 (12968) .33 (.57) .45 (.68) .41 (.63) .49 (.73) .35 (.39) .55 (.96) .06 (.17) .02 (.03) 64.26 (12.54)

SN-I 19 (45) 6945 (12502) .22 (.27) .29 (.34) .22 (.27) .35 (.40) .09 (.01) .48 (.69) .12 (.21) .03 (.03) 27.39 (10.35)

Age I-B 21 (26) 4361 (4343) .41 (.40) .48 (.57) .40 (.46) .56 (.68) .20 (.13) .75 (1.01) .17 (.27) .04 (.06) 15.79 (14.37)

Time frame of PB PB-B 12 (6) 2740 (867) .57 (.61) .64 (.65) .52 (.56) .75 (.73) .32 (.50) .96 (.80) .19 (.09) .06 (.04) 6.10 (27.14)

Note. Statistics for strong attitude-intention strength, older samples and distal time frame within parentheses, weak attitude-intention strength, younger samples and proximal time frame outside parentheses; I = Intention Att = Attitude SN = Subjective Norm B = Behavior PB = Past behavior k = Number of Studies N = Sample size a Average re-weighted correlation corrected for sampling error only b Average re-weighted correlation corrected for sampling error and measurement error ✝ 95% Confidence Interval of average re-weighted correlation corrected for sampling error only ✝✝ 90% Credibility Interval of average re-weighted correlation corrected for sampling and measurement error SD = Standard deviation of mean SE = Standard error of mean


c Percentage of error variance accounted for by statistical artefacts


Figure Caption

Figure 1. Path analysis diagram showing structural relationships between the TPB variables, self-

efficacy and past behavior from corrected correlations derived by meta-analysis.

Note. * p<.01

**p < .05


E3

E4

Attitude

Subjective Norm

PBC

Behavior Intention

Past Behavior

E2 E1 .20*

.37*

.05*

.55*

.39*

.28*

.05*

.23*

.73* .63*

.09*

.22*

Self-Efficacy

E5

.99*

.97*

.92*

.37*

.37*

.51*

.89*E6

.36* .27*

.37*

.45*

.04*

.15*

meta-analysis of the theories of reasoned action and

Documents